pbs-assess/gfmpa

Analysis of closed areas impact on survey indices

gfmpa's Issues

Add column of frequency of model use to Table S1

• add into model fitting return
• pull out into table s1 via values.tex R

Remove 'as-is, where-is' alternative analysis

Decide on delta-Gamma vs. Tweedie model

Both are in there presently

Check if relative error trend relates to original trend

Add traditional design-based variance CIs

Look in SimSurvey as an example

Fig S7 still has QCS + HS joined

Decide if focussing on just model-based indexes

Make plot somehow looking at rank order of set size inside vs. out over time?

TODO after May 3 meeting

• add black line in fig 4?
• add as-is where-is to fig 1
• check... that domain sets are indeed dropped!!
• label strata?
• thin or more alpha on black in stratum figure?
• flip tweedie and delta-gamma order in modelling
• note that in HS: MARE is better with geostatistical:
• make new fig 4 for design based that matches the geostationary version
  (non-free axis and side by side??)
• fig 5: proportions driven by increases or decreases inside vs. outside?
• fig 5: right now, panels b-f are ordered by their position on the x-axis. Consider reordering to make them ordered by the y-axis because they are positioned vertically. This would just involve switching d and e.

Consider design-based estimator

The SYN QCS label gets cut off slightly in Fig 3

Add bootstrap CIs

Need to get the polygons first #8; code otherwise ready.

Examples/stories for species

Darkblotched in WCHG

MPA restriction occasionally misses large sets on west coast:

Results in some years being 'off':

Hence increased MARE, but little impact on trend:

Get stratum polygons and recalculate areas after restriction

Pick a few recent assessments to discuss possible impacts

Ideally ones where different properties (e.g., precision, trend bias) might have had an impact. Growing vs. declining trends etc.

Add figure showing closed areas

Check if Fig 1 (ts) upper limit only affects Petrale Sole in 2007

Remove 2014 from WCHG?

At the very least it should be removed for the design based indices and the comparisons between geostatistical and design based.

Technically could be left in for the geostatistical, although we've tended to just remove it.

Allow simulation in sdmTMB with fitted random field values

Make plot of log centered indexes with regression lines

By species; to verify/understand RE trends.

Also look at regressions of RE itself by species.

Add map beside current map showing the full network

Including as-is, where-is and including those that don't intersect

Add analysis on detection of population changes within MPAs

Add analysis looking at whether these surveys could have detected population changes within (some of) these MPAs.

1. Take fitted models
2. Simulate from them with same non-year fixed effects, same random field parameters, same observation error, various underlying trends

Ask: What is the power to detect various changes in MPAs of various sizes across various species?

• Tame this to something manageable among the various dimensions above.
• Possibly, this should be a separate paper from the bias/precision one.

Add back restricted zones from Gwaii Haanas

Check model 4/5 in Table S1

Look the same right now

Add supplemental analysis resampling additional sets historically

• Sample from existing non-excluded survey sets to match original effort
• Question is how much the metrics are affected by sample size vs. lost area (our guess: precision affected most, others less, others could actually get worse)
• Supplemental figure
• Might need to do 3 or 4 times because which ones sampled will be random!?
• Also possibility of taking these samples from the previously fitted model, but that seems more circular (simulations guaranteed to match fitted model)
• @plthompson feel free to suggest how to do this

Note in fig 3 caption how many spp. ommited from QCS and HS

and change line to use all spp?

Figure out simulation with MPA-year interaction to look at power to detect recovery

Add version of metrics plot (Fig 3) coloured by which model was used